1 /* Copyright (c) 2007-2008 CSIRO 2 Copyright (c) 2007-2009 Xiph.Org Foundation 3 Copyright (c) 2007-2016 Jean-Marc Valin */ 4 /* 5 Redistribution and use in source and binary forms, with or without 6 modification, are permitted provided that the following conditions 7 are met: 8 9 - Redistributions of source code must retain the above copyright 10 notice, this list of conditions and the following disclaimer. 11 12 - Redistributions in binary form must reproduce the above copyright 13 notice, this list of conditions and the following disclaimer in the 14 documentation and/or other materials provided with the distribution. 15 16 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 17 ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 18 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 19 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER 20 OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 21 EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 22 PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 23 PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 24 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 25 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 26 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29 #ifdef HAVE_CONFIG_H 30 #include "config.h" 31 #endif 32 33 #include <xmmintrin.h> 34 #include <emmintrin.h> 35 #include "celt_lpc.h" 36 #include "stack_alloc.h" 37 #include "mathops.h" 38 #include "vq.h" 39 #include "x86cpu.h" 40 41 42 #ifndef FIXED_POINT 43 44 opus_val16 op_pvq_search_sse2(celt_norm *_X, int *iy, int K, int N, int arch) 45 { 46 int i, j; 47 int pulsesLeft; 48 float xy, yy; 49 VARDECL(celt_norm, y); 50 VARDECL(celt_norm, X); 51 VARDECL(float, signy); 52 __m128 signmask; 53 __m128 sums; 54 __m128i fours; 55 SAVE_STACK; 56 57 (void)arch; 58 /* All bits set to zero, except for the sign bit. */ 59 signmask = _mm_set_ps1(-0.f); 60 fours = _mm_set_epi32(4, 4, 4, 4); 61 ALLOC(y, N+3, celt_norm); 62 ALLOC(X, N+3, celt_norm); 63 ALLOC(signy, N+3, float); 64 65 OPUS_COPY(X, _X, N); 66 X[N] = X[N+1] = X[N+2] = 0; 67 sums = _mm_setzero_ps(); 68 for (j=0;j<N;j+=4) 69 { 70 __m128 x4, s4; 71 x4 = _mm_loadu_ps(&X[j]); 72 s4 = _mm_cmplt_ps(x4, _mm_setzero_ps()); 73 /* Get rid of the sign */ 74 x4 = _mm_andnot_ps(signmask, x4); 75 sums = _mm_add_ps(sums, x4); 76 /* Clear y and iy in case we don't do the projection. */ 77 _mm_storeu_ps(&y[j], _mm_setzero_ps()); 78 _mm_storeu_si128((__m128i*)&iy[j], _mm_setzero_si128()); 79 _mm_storeu_ps(&X[j], x4); 80 _mm_storeu_ps(&signy[j], s4); 81 } 82 sums = _mm_add_ps(sums, _mm_shuffle_ps(sums, sums, _MM_SHUFFLE(1, 0, 3, 2))); 83 sums = _mm_add_ps(sums, _mm_shuffle_ps(sums, sums, _MM_SHUFFLE(2, 3, 0, 1))); 84 85 xy = yy = 0; 86 87 pulsesLeft = K; 88 89 /* Do a pre-search by projecting on the pyramid */ 90 if (K > (N>>1)) 91 { 92 __m128i pulses_sum; 93 __m128 yy4, xy4; 94 __m128 rcp4; 95 opus_val32 sum = _mm_cvtss_f32(sums); 96 /* If X is too small, just replace it with a pulse at 0 */ 97 /* Prevents infinities and NaNs from causing too many pulses 98 to be allocated. 64 is an approximation of infinity here. */ 99 if (!(sum > EPSILON && sum < 64)) 100 { 101 X[0] = QCONST16(1.f,14); 102 j=1; do 103 X[j]=0; 104 while (++j<N); 105 sums = _mm_set_ps1(1.f); 106 } 107 /* Using K+e with e < 1 guarantees we cannot get more than K pulses. */ 108 rcp4 = _mm_mul_ps(_mm_set_ps1((float)(K+.8)), _mm_rcp_ps(sums)); 109 xy4 = yy4 = _mm_setzero_ps(); 110 pulses_sum = _mm_setzero_si128(); 111 for (j=0;j<N;j+=4) 112 { 113 __m128 rx4, x4, y4; 114 __m128i iy4; 115 x4 = _mm_loadu_ps(&X[j]); 116 rx4 = _mm_mul_ps(x4, rcp4); 117 iy4 = _mm_cvttps_epi32(rx4); 118 pulses_sum = _mm_add_epi32(pulses_sum, iy4); 119 _mm_storeu_si128((__m128i*)&iy[j], iy4); 120 y4 = _mm_cvtepi32_ps(iy4); 121 xy4 = _mm_add_ps(xy4, _mm_mul_ps(x4, y4)); 122 yy4 = _mm_add_ps(yy4, _mm_mul_ps(y4, y4)); 123 /* double the y[] vector so we don't have to do it in the search loop. */ 124 _mm_storeu_ps(&y[j], _mm_add_ps(y4, y4)); 125 } 126 pulses_sum = _mm_add_epi32(pulses_sum, _mm_shuffle_epi32(pulses_sum, _MM_SHUFFLE(1, 0, 3, 2))); 127 pulses_sum = _mm_add_epi32(pulses_sum, _mm_shuffle_epi32(pulses_sum, _MM_SHUFFLE(2, 3, 0, 1))); 128 pulsesLeft -= _mm_cvtsi128_si32(pulses_sum); 129 xy4 = _mm_add_ps(xy4, _mm_shuffle_ps(xy4, xy4, _MM_SHUFFLE(1, 0, 3, 2))); 130 xy4 = _mm_add_ps(xy4, _mm_shuffle_ps(xy4, xy4, _MM_SHUFFLE(2, 3, 0, 1))); 131 xy = _mm_cvtss_f32(xy4); 132 yy4 = _mm_add_ps(yy4, _mm_shuffle_ps(yy4, yy4, _MM_SHUFFLE(1, 0, 3, 2))); 133 yy4 = _mm_add_ps(yy4, _mm_shuffle_ps(yy4, yy4, _MM_SHUFFLE(2, 3, 0, 1))); 134 yy = _mm_cvtss_f32(yy4); 135 } 136 X[N] = X[N+1] = X[N+2] = -100; 137 y[N] = y[N+1] = y[N+2] = 100; 138 celt_assert2(pulsesLeft>=0, "Allocated too many pulses in the quick pass"); 139 140 /* This should never happen, but just in case it does (e.g. on silence) 141 we fill the first bin with pulses. */ 142 if (pulsesLeft > N+3) 143 { 144 opus_val16 tmp = (opus_val16)pulsesLeft; 145 yy = MAC16_16(yy, tmp, tmp); 146 yy = MAC16_16(yy, tmp, y[0]); 147 iy[0] += pulsesLeft; 148 pulsesLeft=0; 149 } 150 151 for (i=0;i<pulsesLeft;i++) 152 { 153 int best_id; 154 __m128 xy4, yy4; 155 __m128 max, max2; 156 __m128i count; 157 __m128i pos; 158 /* The squared magnitude term gets added anyway, so we might as well 159 add it outside the loop */ 160 yy = ADD16(yy, 1); 161 xy4 = _mm_load1_ps(&xy); 162 yy4 = _mm_load1_ps(&yy); 163 max = _mm_setzero_ps(); 164 pos = _mm_setzero_si128(); 165 count = _mm_set_epi32(3, 2, 1, 0); 166 for (j=0;j<N;j+=4) 167 { 168 __m128 x4, y4, r4; 169 x4 = _mm_loadu_ps(&X[j]); 170 y4 = _mm_loadu_ps(&y[j]); 171 x4 = _mm_add_ps(x4, xy4); 172 y4 = _mm_add_ps(y4, yy4); 173 y4 = _mm_rsqrt_ps(y4); 174 r4 = _mm_mul_ps(x4, y4); 175 /* Update the index of the max. */ 176 pos = _mm_max_epi16(pos, _mm_and_si128(count, _mm_castps_si128(_mm_cmpgt_ps(r4, max)))); 177 /* Update the max. */ 178 max = _mm_max_ps(max, r4); 179 /* Update the indices (+4) */ 180 count = _mm_add_epi32(count, fours); 181 } 182 /* Horizontal max */ 183 max2 = _mm_max_ps(max, _mm_shuffle_ps(max, max, _MM_SHUFFLE(1, 0, 3, 2))); 184 max2 = _mm_max_ps(max2, _mm_shuffle_ps(max2, max2, _MM_SHUFFLE(2, 3, 0, 1))); 185 /* Now that max2 contains the max at all positions, look at which value(s) of the 186 partial max is equal to the global max. */ 187 pos = _mm_and_si128(pos, _mm_castps_si128(_mm_cmpeq_ps(max, max2))); 188 pos = _mm_max_epi16(pos, _mm_unpackhi_epi64(pos, pos)); 189 pos = _mm_max_epi16(pos, _mm_shufflelo_epi16(pos, _MM_SHUFFLE(1, 0, 3, 2))); 190 best_id = _mm_cvtsi128_si32(pos); 191 192 /* Updating the sums of the new pulse(s) */ 193 xy = ADD32(xy, EXTEND32(X[best_id])); 194 /* We're multiplying y[j] by two so we don't have to do it here */ 195 yy = ADD16(yy, y[best_id]); 196 197 /* Only now that we've made the final choice, update y/iy */ 198 /* Multiplying y[j] by 2 so we don't have to do it everywhere else */ 199 y[best_id] += 2; 200 iy[best_id]++; 201 } 202 203 /* Put the original sign back */ 204 for (j=0;j<N;j+=4) 205 { 206 __m128i y4; 207 __m128i s4; 208 y4 = _mm_loadu_si128((__m128i*)&iy[j]); 209 s4 = _mm_castps_si128(_mm_loadu_ps(&signy[j])); 210 y4 = _mm_xor_si128(_mm_add_epi32(y4, s4), s4); 211 _mm_storeu_si128((__m128i*)&iy[j], y4); 212 } 213 RESTORE_STACK; 214 return yy; 215 } 216 217 #endif 218
